Using modeling results

Sometimes the outcome is not a natural feature of the model, e.g. costs in an epidemic

inputs an outcome, or feature of the model output, you want to control optimal control theory dealing with uncertainty

When are predictions (post-dictions) “causal”?

Interpretation and communication

Show model fit, but also compute diagnostics. Show several models, possibly nested.

Be careful about counterfactuals. Explain when you are projecting beyond support of the data.